Erik Kusch, PhD Student

Department of Biology

Section for Ecoinformatics & Biodiversity

Center for Biodiversity Dynamics in a Changing World (BIOCHANGE)

Aarhus University

05/02/2021

[Study Group] Bayesian Statistics with the Rethinking Material 1

05/02/2021

[Study Group] Bayesian Statistics with the Rethinking Material 2

Models need to straddle the line between being too

simplistic and being overly complex.

05/02/2021

[Study Group] Bayesian Statistics with the Rethinking Material 3

05/02/2021

[Study Group] Bayesian Statistics with the Rethinking Material 4

Models ought to be regularised (tuned for

desired complexity-accuracy)!

05/02/2021

[Study Group] Bayesian Statistics with the Rethinking Material 5

We can compare models without knowing the absolute truth.

05/02/2021

[Study Group] Bayesian Statistics with the Rethinking Material 6

Difference in

log-score

between

training and

test data

reveals

overfitting

05/02/2021

[Study Group] Bayesian Statistics with the Rethinking Material 7

Regularised priors fair worse in-

sample, but better out-of-sample.

Regularisation effects diminish with increasing sample sizes.

05/02/2021

[Study Group] Bayesian Statistics with the Rethinking Material 8

This is the model we would want to select

because it shows the treatment effect

Model Selection can lead to mis-

identification of causal pathways.